From yurii.aulchenko at gmail.com  Fri Jun 20 11:03:18 2014
From: yurii.aulchenko at gmail.com (Yury Aulchenko)
Date: Fri, 20 Jun 2014 11:03:18 +0200
Subject: [GenABEL-dev] Fwd: Missing file for unit test in GenABEL
References: <20140620084443.GA17492@an3as.eu>
Message-ID: <E4124229-29F5-4DEC-ADB6-91038E9F9FB9@gmail.com>

FYI

Begin forwarded message:

> From: Andreas Tille <andreas at an3as.eu>
> Subject: Missing file for unit test in GenABEL
> Date: June 20, 2014 at 10:44:43 GMT+2
> To: Yurii Aulchenko <yurii at bionet.nsc.ru>, Debian Med Packaging Team <debian-med-packaging at lists.alioth.debian.org>
> 
> Hi Yurii,
> 
> I'm trying to update the Debian package of GenABEL.  Since some time
> there is an effort to automatically run unit tests of software if
> available.  Since I noticed that GenABEL comes with unit tests I
> tried
> 
> $ make test
> export RCMDCHECK=FALSE;\
>        cd ../../tests;\
>        R --vanilla --slave < doRUnit.R
> /bin/sh: 2: cd: can't cd to ../../tests
> /bin/sh: 3: cannot open doRUnit.R: No such file
> make: *** [test] Error 2
> 
> 
> As you can see the file doRUnit.R is missing.  It would be great if you
> could include this file into the source diustribution to make sure we
> can reproduce your exact test procedure in the Debian package.
> 
> Kind regards and thanks for providing GenABEL as Free Software
> 
>      Andreas.
> 
> -- 
> http://fam-tille.de

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140620/19945ca5/attachment.html>

From lennart at karssen.org  Fri Jun 20 12:18:04 2014
From: lennart at karssen.org (L.C. Karssen)
Date: Fri, 20 Jun 2014 12:18:04 +0200
Subject: [GenABEL-dev] Why is doRUnit.R removed from the final package?
Message-ID: <53A40A5C.6000903@karssen.org>

Dear list,

I just noticed a commit in the Debian packaging system for the Debian
package of GenABEL (r-cran-genabel) [1]. The packager (Andreas Tille)
wrote in the log that a file is missing (tests/doRUnit.R). It turns out
that this file is removed in our makedistrib_GenABEL.sh script [2]. Does
anyone remember why this is done?


Thanks,

Lennart.


[1] http://anonscm.debian.org/viewvc/debian-med?view=revision&revision=17252
[2]
https://r-forge.r-project.org/scm/viewvc.php/pkg/GenABEL-general/distrib_scripts/makedistrib_GenABEL.sh?view=markup&revision=1684&root=genabel
-- 
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 213 bytes
Desc: OpenPGP digital signature
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140620/bbec5a04/attachment.sig>

From yurii.aulchenko at gmail.com  Fri Jun 20 12:25:55 2014
From: yurii.aulchenko at gmail.com (Yury Aulchenko)
Date: Fri, 20 Jun 2014 12:25:55 +0200
Subject: [GenABEL-dev] Why is doRUnit.R removed from the final package?
In-Reply-To: <53A40A5C.6000903@karssen.org>
References: <53A40A5C.6000903@karssen.org>
Message-ID: <8F9D92B7-2F93-4587-B4DE-436299A5A897@gmail.com>

because CRAN requested NOT to include unit tests into distrib - my unerstanding was that it takes to long + tests are not stable; again, for the latter my understanding was that it is not specifically GenABEL, it is something general

Lennart, so you think we should address Andreas to our SVN?

Yurii


On Jun 20, 2014, at 12:18, L.C. Karssen <lennart at karssen.org> wrote:

> Dear list,
> 
> I just noticed a commit in the Debian packaging system for the Debian
> package of GenABEL (r-cran-genabel) [1]. The packager (Andreas Tille)
> wrote in the log that a file is missing (tests/doRUnit.R). It turns out
> that this file is removed in our makedistrib_GenABEL.sh script [2]. Does
> anyone remember why this is done?
> 
> 
> Thanks,
> 
> Lennart.
> 
> 
> [1] http://anonscm.debian.org/viewvc/debian-med?view=revision&revision=17252
> [2]
> https://r-forge.r-project.org/scm/viewvc.php/pkg/GenABEL-general/distrib_scripts/makedistrib_GenABEL.sh?view=markup&revision=1684&root=genabel
> -- 
> *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
> L.C. Karssen
> Utrecht
> The Netherlands
> 
> lennart at karssen.org
> http://blog.karssen.org
> GPG key ID: A88F554A
> -*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-
> 
> _______________________________________________
> genabel-devel mailing list
> genabel-devel at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel


From lennart at karssen.org  Fri Jun 20 14:13:32 2014
From: lennart at karssen.org (L.C. Karssen)
Date: Fri, 20 Jun 2014 14:13:32 +0200
Subject: [GenABEL-dev] Why is doRUnit.R removed from the final package?
In-Reply-To: <8F9D92B7-2F93-4587-B4DE-436299A5A897@gmail.com>
References: <53A40A5C.6000903@karssen.org>
 <8F9D92B7-2F93-4587-B4DE-436299A5A897@gmail.com>
Message-ID: <53A4256C.4070406@karssen.org>

Hi Yurii,

I see Andreas already contacted you before I noticed his commit.


On 20-06-14 12:25, Yury Aulchenko wrote:
> because CRAN requested NOT to include unit tests into distrib - my
> unerstanding was that it takes to long + tests are not stable; again,
> for the latter my understanding was that it is not specifically GenABEL,
> it is something general

I quickly checked a few packages on CRAN and I'm not sure about what the
results mean:
- Rcpp: has doRunit.R
- MASS: has tests, but no doRUnit.R (maybe not using RUnit?)
- ggplot2: has tests, but no doRUnit.R (maybe not using RUnit?)
- HMisc: has tests, but no doRUnit.R (maybe not using RUnit?)

As far as I can see the "Writing R Extensions" manual doesn't mention
anything about doRUnit.R specifically.


> 
> Lennart, so you think we should address Andreas to our SVN?

I guess you meant "add" instead of "address"? If so, then no, that
wasn't my idea. Andreas is one of the main forces behind the Debian Med
team, which focusses on making Debian packages for Medical/Life Sciences
software. I don't think he is interested in participating in the
development of GenABEL specifically (but we could ask).


Lennart.

> 
> Yurii
> 
> 
> On Jun 20, 2014, at 12:18, L.C. Karssen <lennart at karssen.org> wrote:
> 
>> Dear list,
>>
>> I just noticed a commit in the Debian packaging system for the Debian
>> package of GenABEL (r-cran-genabel) [1]. The packager (Andreas Tille)
>> wrote in the log that a file is missing (tests/doRUnit.R). It turns out
>> that this file is removed in our makedistrib_GenABEL.sh script [2]. Does
>> anyone remember why this is done?
>>
>>
>> Thanks,
>>
>> Lennart.
>>
>>
>> [1] http://anonscm.debian.org/viewvc/debian-med?view=revision&revision=17252
>> [2]
>> https://r-forge.r-project.org/scm/viewvc.php/pkg/GenABEL-general/distrib_scripts/makedistrib_GenABEL.sh?view=markup&revision=1684&root=genabel
>> -- 
>> *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
>> L.C. Karssen
>> Utrecht
>> The Netherlands
>>
>> lennart at karssen.org
>> http://blog.karssen.org
>> GPG key ID: A88F554A
>> -*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-
>>
>> _______________________________________________
>> genabel-devel mailing list
>> genabel-devel at lists.r-forge.r-project.org
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
> 

-- 
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 213 bytes
Desc: OpenPGP digital signature
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140620/9f5bd66f/attachment.sig>

From lennart at karssen.org  Mon Jun 23 14:32:54 2014
From: lennart at karssen.org (L.C. Karssen)
Date: Mon, 23 Jun 2014 14:32:54 +0200
Subject: [GenABEL-dev] [Genabel-commits] r1748 - in pkg/OmicABELnoMM: .
	src tests
In-Reply-To: <538482AF.2060507@karssen.org>
References: <20140527120853.AB60C1873C7@r-forge.r-project.org>
 <538482AF.2060507@karssen.org>
Message-ID: <53A81E76.7080306@karssen.org>

Hi Alvaro,

I just tried to build OmicABELnoMM and it misses the file
tests/Makefile.am in SVN.
It looks like you have forgotten to add tests/Makefile.am to SVN, right?
I must have overlooked the missing Makefile.am file when reviewing your
commit (see mail message below).


Thanks,

Lennart.

On 27-05-14 14:18, L.C. Karssen wrote:
> Hi Alvaro,
> 
> On 27-05-14 14:08, noreply at r-forge.r-project.org wrote:
>> Author: afrank
>> Date: 2014-05-27 14:08:53 +0200 (Tue, 27 May 2014)
>> New Revision: 1748
>>
>> Modified:
>>    pkg/OmicABELnoMM/Makefile.am
>>    pkg/OmicABELnoMM/configure.ac
>>    pkg/OmicABELnoMM/src/Algorithm.cpp
>>    pkg/OmicABELnoMM/tests/Makefile
>>    pkg/OmicABELnoMM/tests/test.cpp
>> Log:
>> Automake integration of tests now runs them using make check. Tests are also compiled along with the normal executable.
> 
> 
> That sounds good!
> 
>>
>>
>> Modified: pkg/OmicABELnoMM/tests/Makefile
>> ===================================================================
> 
> Now that you have a Makefile.am, the Makefile itself can be removed from
> SVN.
> 
> 
> Thanks a lot!
> 
> Lennart.
> 
>>
>> To get the complete diff run:
>>     svnlook diff /svnroot/genabel -r 1748
>> _______________________________________________
>> Genabel-commits mailing list
>> Genabel-commits at lists.r-forge.r-project.org
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-commits
>>
> 

-- 
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 213 bytes
Desc: OpenPGP digital signature
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140623/80792045/attachment.sig>

From alvaro.frank at rwth-aachen.de  Mon Jun 23 14:52:42 2014
From: alvaro.frank at rwth-aachen.de (Frank, Alvaro Jesus)
Date: Mon, 23 Jun 2014 12:52:42 +0000
Subject: [GenABEL-dev] [Genabel-commits] r1748 - in pkg/OmicABELnoMM:
	.	src tests
In-Reply-To: <53A81E76.7080306@karssen.org>
References: <20140527120853.AB60C1873C7@r-forge.r-project.org>
 <538482AF.2060507@karssen.org>,<53A81E76.7080306@karssen.org>
Message-ID: <244CF001646FF74FB34F372310A332C57BBD78@MBX5.rwth-ad.de>

Hi Lennart, 

I will fix it. At the moment I am working on doing efficient calculations of t-score, R^2 and p-value. This are as expensive as the calculation of the slopes from regression themselves so it has been quite time consuming in the design phase. I will have some prototype soon.

-Alvaro

________________________________________
From: genabel-devel-bounces at lists.r-forge.r-project.org [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of L.C. Karssen [lennart at karssen.org]
Sent: Monday, June 23, 2014 2:32 PM
To: genabel-devel at lists.r-forge.r-project.org
Subject: Re: [GenABEL-dev] [Genabel-commits] r1748 - in pkg/OmicABELnoMM: .     src tests

Hi Alvaro,

I just tried to build OmicABELnoMM and it misses the file
tests/Makefile.am in SVN.
It looks like you have forgotten to add tests/Makefile.am to SVN, right?
I must have overlooked the missing Makefile.am file when reviewing your
commit (see mail message below).


Thanks,

Lennart.

On 27-05-14 14:18, L.C. Karssen wrote:
> Hi Alvaro,
>
> On 27-05-14 14:08, noreply at r-forge.r-project.org wrote:
>> Author: afrank
>> Date: 2014-05-27 14:08:53 +0200 (Tue, 27 May 2014)
>> New Revision: 1748
>>
>> Modified:
>>    pkg/OmicABELnoMM/Makefile.am
>>    pkg/OmicABELnoMM/configure.ac
>>    pkg/OmicABELnoMM/src/Algorithm.cpp
>>    pkg/OmicABELnoMM/tests/Makefile
>>    pkg/OmicABELnoMM/tests/test.cpp
>> Log:
>> Automake integration of tests now runs them using make check. Tests are also compiled along with the normal executable.
>
>
> That sounds good!
>
>>
>>
>> Modified: pkg/OmicABELnoMM/tests/Makefile
>> ===================================================================
>
> Now that you have a Makefile.am, the Makefile itself can be removed from
> SVN.
>
>
> Thanks a lot!
>
> Lennart.
>
>>
>> To get the complete diff run:
>>     svnlook diff /svnroot/genabel -r 1748
>> _______________________________________________
>> Genabel-commits mailing list
>> Genabel-commits at lists.r-forge.r-project.org
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-commits
>>
>

--
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-


From alvaro.frank at rwth-aachen.de  Mon Jun 23 15:21:36 2014
From: alvaro.frank at rwth-aachen.de (Frank, Alvaro Jesus)
Date: Mon, 23 Jun 2014 13:21:36 +0000
Subject: [GenABEL-dev] P-values
Message-ID: <244CF001646FF74FB34F372310A332C57BBD8F@MBX5.rwth-ad.de>

Hi all,

I have been trying to figure out an efficient way to calculate p-values, and it seems that I managed to come to an efficient compromise between speed an accuracy. 99% of the regressions will yield a non significant p value, that is to say p > 0.05 or even conservatively 0.1 . It is easy to know apriori if the p-value is significant, by looking at the t-score where it originates from. For any t-score < 1.28 the pvalue will not go below 0.1 for a t-distribution or normal distribution. In this cases (9X%) an aproximation with an error of 10^-(4~5) of the p-value is enough. This calculation is efficient involving only a quadratic polynomial to be approximated. For possible significant p-values, with t-score > 1.28 a proper calculation of the p-value can be done. Note that a p-value calculation involves approximating the integral of the distribution used, either t-students (n<1000?) or normal distribution.

How are the different genabel packages handling this at the moment? This is a speedup that can be applied to any p-value calculation.

Plotted Error between 1-ncdf(x)  and polynomial approx:

http://www.wolframalpha.com/input/?i=y+%3D+%28%281%2F2+-+1%2F2*erf%28x%2Fsqrt%282%29%29%29+-+%281%2F2-%280.1*x*%284.4-x%29%29%29%29+from+0+to+1.28


-Alvaro
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140623/bd5e19dd/attachment.html>

From lennart at karssen.org  Mon Jun 23 15:22:16 2014
From: lennart at karssen.org (L.C. Karssen)
Date: Mon, 23 Jun 2014 15:22:16 +0200
Subject: [GenABEL-dev] [Genabel-commits] r1748 - in pkg/OmicABELnoMM: .
 src tests
In-Reply-To: <244CF001646FF74FB34F372310A332C57BBD78@MBX5.rwth-ad.de>
References: <20140527120853.AB60C1873C7@r-forge.r-project.org>
 <538482AF.2060507@karssen.org>, <53A81E76.7080306@karssen.org>
 <244CF001646FF74FB34F372310A332C57BBD78@MBX5.rwth-ad.de>
Message-ID: <53A82A08.7030005@karssen.org>

Hi Alvaro,

On 23-06-14 14:52, Frank, Alvaro Jesus wrote:
> Hi Lennart, 
> 
> I will fix it.

Thanks!

>  At the moment I am working on doing efficient calculations of t-score, R^2 and p-value. This are as expensive as the calculation of the slopes from regression themselves so it has been quite time consuming in the design phase. I will have some prototype soon. 

Sounds cool!

By the way, I've been rather busy lately, but I've seen your messages
about big data files on this list. Since your mails are quite
substantial I need some longer stretch of time to think them over and
compose a meaningful reply. I intend to do so in the coming weeks.


Best,

Lennart.

> 
> -Alvaro
> 
> ________________________________________
> From: genabel-devel-bounces at lists.r-forge.r-project.org [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of L.C. Karssen [lennart at karssen.org]
> Sent: Monday, June 23, 2014 2:32 PM
> To: genabel-devel at lists.r-forge.r-project.org
> Subject: Re: [GenABEL-dev] [Genabel-commits] r1748 - in pkg/OmicABELnoMM: .     src tests
> 
> Hi Alvaro,
> 
> I just tried to build OmicABELnoMM and it misses the file
> tests/Makefile.am in SVN.
> It looks like you have forgotten to add tests/Makefile.am to SVN, right?
> I must have overlooked the missing Makefile.am file when reviewing your
> commit (see mail message below).
> 
> 
> Thanks,
> 
> Lennart.
> 
> On 27-05-14 14:18, L.C. Karssen wrote:
>> Hi Alvaro,
>>
>> On 27-05-14 14:08, noreply at r-forge.r-project.org wrote:
>>> Author: afrank
>>> Date: 2014-05-27 14:08:53 +0200 (Tue, 27 May 2014)
>>> New Revision: 1748
>>>
>>> Modified:
>>>    pkg/OmicABELnoMM/Makefile.am
>>>    pkg/OmicABELnoMM/configure.ac
>>>    pkg/OmicABELnoMM/src/Algorithm.cpp
>>>    pkg/OmicABELnoMM/tests/Makefile
>>>    pkg/OmicABELnoMM/tests/test.cpp
>>> Log:
>>> Automake integration of tests now runs them using make check. Tests are also compiled along with the normal executable.
>>
>>
>> That sounds good!
>>
>>>
>>>
>>> Modified: pkg/OmicABELnoMM/tests/Makefile
>>> ===================================================================
>>
>> Now that you have a Makefile.am, the Makefile itself can be removed from
>> SVN.
>>
>>
>> Thanks a lot!
>>
>> Lennart.
>>
>>>
>>> To get the complete diff run:
>>>     svnlook diff /svnroot/genabel -r 1748
>>> _______________________________________________
>>> Genabel-commits mailing list
>>> Genabel-commits at lists.r-forge.r-project.org
>>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-commits
>>>
>>
> 
> --
> *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
> L.C. Karssen
> Utrecht
> The Netherlands
> 
> lennart at karssen.org
> http://blog.karssen.org
> GPG key ID: A88F554A
> -*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-
> 

-- 
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 213 bytes
Desc: OpenPGP digital signature
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140623/e8fef11b/attachment.sig>

From lennart at karssen.org  Mon Jun 23 15:51:21 2014
From: lennart at karssen.org (L.C. Karssen)
Date: Mon, 23 Jun 2014 15:51:21 +0200
Subject: [GenABEL-dev] P-values
In-Reply-To: <244CF001646FF74FB34F372310A332C57BBD8F@MBX5.rwth-ad.de>
References: <244CF001646FF74FB34F372310A332C57BBD8F@MBX5.rwth-ad.de>
Message-ID: <53A830D9.8020008@karssen.org>

Hi Alvaro,

On 23-06-14 15:21, Frank, Alvaro Jesus wrote:
> Hi all,
> 
> I have been trying to figure out an efficient way to calculate p-values,
> and it seems that I managed to come to an efficient compromise between
> speed an accuracy.

That sounds cool. An (unnecessary?) word of caution though: I don't know
how much research you have done on this before starting, but make sure
to look at existing implementations of p-value calculation. Given that
in genetics p-values can become extremely small (e.g. 1e-100 or smaller)
numerical problems abound.
See also http://dx.doi.org/10.1016/j.csda.2008.11.028

> 99% of the regressions will yield a non significant p
> value, that is to say p > 0.05 or even conservatively 0.1 . It is easy
> to know apriori if the p-value is significant, by looking at the t-score
> where it originates from. For any t-score < 1.28 the pvalue will not go
> below 0.1 for a t-distribution or normal distribution. In this cases
> (9X%) an aproximation with an error of 10^-(4~5) of the p-value is

Do you mean an absolute or relative error of 1e-4 here?

> enough. This calculation is efficient involving only a quadratic
> polynomial to be approximated. For possible significant p-values, with
> t-score > 1.28 a proper calculation of the p-value can be done. 

I like the idea of setting a threshold for t, below which doing slow but
accurate calculations make no sense. If that gives a considerable speed
up, go for it!
I would make the actual threshold a user-definable option, by the way.
There may be use cases where people want to have accurate p-values even
for non-significant cases.

> Note
> that a p-value calculation involves approximating the integral of the
> distribution used, either t-students (n<1000?) or normal distribution.

Be sure to check out existing implementations (e.g. in Boost) on how
integration is being handled. Somewhere at the back of my mind the term
'incomplete Gamma function' comes to mind.

> 
> How are the different genabel packages handling this at the moment? This
> is a speedup that can be applied to any p-value calculation.

The R-based packages will most likely use the functions provided by R,
e.g. pchisq() for calculating the p-value of a chi^2 statistic.
In ProbABEL we are now implementing the calculation of p-values, but the
actual calculation is being left to the Boost libraries. Currently we
calculate the p-values irrespective of the value of the statistic. I'll
profile the code to see how much time this takes and whether it makes
sense to only calculate p-values for statistics with a value above a
certain threshold.

> 
> Plotted Error between 1-ncdf(x)  and polynomial approx:
> 
> http://www.wolframalpha.com/input/?i=y+%3D+%28%281%2F2+-+1%2F2*erf%28x%2Fsqrt%282%29%29%29+-+%281%2F2-%280.1*x*%284.4-x%29%29%29%29+from+0+to+1.28
> 

ah, that leads me to believe the answer to my question above is that you
mean an absolute error of ~1e-4, correct?


Best,

Lennart.

> 
> -Alvaro
> 
> 
> _______________________________________________
> genabel-devel mailing list
> genabel-devel at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
> 

-- 
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 213 bytes
Desc: OpenPGP digital signature
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140623/b24c60e2/attachment.sig>

From alvaro.frank at rwth-aachen.de  Mon Jun 23 16:02:21 2014
From: alvaro.frank at rwth-aachen.de (Frank, Alvaro Jesus)
Date: Mon, 23 Jun 2014 14:02:21 +0000
Subject: [GenABEL-dev] P-values
In-Reply-To: <53A830D9.8020008@karssen.org>
References: <244CF001646FF74FB34F372310A332C57BBD8F@MBX5.rwth-ad.de>,
 <53A830D9.8020008@karssen.org>
Message-ID: <244CF001646FF74FB34F372310A332C57BBDD8@MBX5.rwth-ad.de>

That sounds cool. An (unnecessary?) word of caution though: I don't know
how much research you have done on this before starting, but make sure
to look at existing implementations of p-value calculation. Given that
in genetics p-values can become extremely small (e.g. 1e-100 or smaller)
numerical problems abound.
See also http://dx.doi.org/10.1016/j.csda.2008.11.028

Be sure to check out existing implementations (e.g. in Boost) on how
integration is being handled. Somewhere at the back of my mind the term
'incomplete Gamma function' comes to mind.

Such extreme values are directly related to the t-score, and it starts being a problem at around 20+.
For those cases I am set to use arbitrary precise functions as in: http://www.mpfr.org/mpfr-current/mpfr.html

Which leads me to the critic that if people only report p-values as their work, perhaps there is a cultural and scientific problem with the workflow and conceptual basis. But if they want their 3.5*10^-300 I will calculate it.

I would make the actual threshold a user-definable option, by the way.
There may be use cases where people want to have accurate p-values even
for non-significant cases.
I was thinking the same. From Local workflows here at Helmholtz I see that any insignificant value gets ignored completely unless it comes to doing Manhattan plots. So another setting would be to not calculate or store at all anything for non significant t-scores, which is helpfull during early discovery of significant signals. For Final calculations for plotting and reporting, the setting can then be changed to enable those. This effectivly removes the "Big data" problem, since 99% of the data is meaningless to determining whether the slope is significant.

In ProbABEL we are now implementing the calculation of p-values, but the
actual calculation is being left to the Boost libraries. Currently we
calculate the p-values irrespective of the value of the statistic. I'll
profile the code to see how much time this takes and whether it makes
sense to only calculate p-values for statistics with a value above a
certain threshold.

I was considering boost, if your profiling is good, perhaps boost is doing the threshold already. Let me know to see if I should switch over to boost too.

-A Frank
________________________________________
From: genabel-devel-bounces at lists.r-forge.r-project.org [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of L.C. Karssen [lennart at karssen.org]
Sent: Monday, June 23, 2014 3:51 PM
To: genabel-devel at lists.r-forge.r-project.org
Subject: Re: [GenABEL-dev] P-values

Hi Alvaro,

On 23-06-14 15:21, Frank, Alvaro Jesus wrote:
> Hi all,
>
> I have been trying to figure out an efficient way to calculate p-values,
> and it seems that I managed to come to an efficient compromise between
> speed an accuracy.

That sounds cool. An (unnecessary?) word of caution though: I don't know
how much research you have done on this before starting, but make sure
to look at existing implementations of p-value calculation. Given that
in genetics p-values can become extremely small (e.g. 1e-100 or smaller)
numerical problems abound.
See also http://dx.doi.org/10.1016/j.csda.2008.11.028

> 99% of the regressions will yield a non significant p
> value, that is to say p > 0.05 or even conservatively 0.1 . It is easy
> to know apriori if the p-value is significant, by looking at the t-score
> where it originates from. For any t-score < 1.28 the pvalue will not go
> below 0.1 for a t-distribution or normal distribution. In this cases
> (9X%) an aproximation with an error of 10^-(4~5) of the p-value is

Do you mean an absolute or relative error of 1e-4 here?

> enough. This calculation is efficient involving only a quadratic
> polynomial to be approximated. For possible significant p-values, with
> t-score > 1.28 a proper calculation of the p-value can be done.

I like the idea of setting a threshold for t, below which doing slow but
accurate calculations make no sense. If that gives a considerable speed
up, go for it!
I would make the actual threshold a user-definable option, by the way.
There may be use cases where people want to have accurate p-values even
for non-significant cases.

> Note
> that a p-value calculation involves approximating the integral of the
> distribution used, either t-students (n<1000?) or normal distribution.

Be sure to check out existing implementations (e.g. in Boost) on how
integration is being handled. Somewhere at the back of my mind the term
'incomplete Gamma function' comes to mind.

>
> How are the different genabel packages handling this at the moment? This
> is a speedup that can be applied to any p-value calculation.

The R-based packages will most likely use the functions provided by R,
e.g. pchisq() for calculating the p-value of a chi^2 statistic.
In ProbABEL we are now implementing the calculation of p-values, but the
actual calculation is being left to the Boost libraries. Currently we
calculate the p-values irrespective of the value of the statistic. I'll
profile the code to see how much time this takes and whether it makes
sense to only calculate p-values for statistics with a value above a
certain threshold.

>
> Plotted Error between 1-ncdf(x) and polynomial approx:
>
> http://www.wolframalpha.com/input/?i=y+%3D+%28%281%2F2+-+1%2F2*erf%28x%2Fsqrt%282%29%29%29+-+%281%2F2-%280.1*x*%284.4-x%29%29%29%29+from+0+to+1.28
>

ah, that leads me to believe the answer to my question above is that you
mean an absolute error of ~1e-4, correct?


Best,

Lennart.

>
> -Alvaro
>
>
> _______________________________________________
> genabel-devel mailing list
> genabel-devel at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
>

--
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140623/aebf06a1/attachment-0001.html>

From alvaro.frank at rwth-aachen.de  Mon Jun 23 16:38:43 2014
From: alvaro.frank at rwth-aachen.de (Frank, Alvaro Jesus)
Date: Mon, 23 Jun 2014 14:38:43 +0000
Subject: [GenABEL-dev] P-values
In-Reply-To: <244CF001646FF74FB34F372310A332C57BBDD8@MBX5.rwth-ad.de>
References: <244CF001646FF74FB34F372310A332C57BBD8F@MBX5.rwth-ad.de>,
 <53A830D9.8020008@karssen.org>,
 <244CF001646FF74FB34F372310A332C57BBDD8@MBX5.rwth-ad.de>
Message-ID: <244CF001646FF74FB34F372310A332C57BBDF2@MBX5.rwth-ad.de>

Sorry for the p-value spam, but...

On Another note, it is important to note that a t-score is equivalent (in the 2 tailed case or a well defined head/tail) to any p-value since there is a 1-1 correspondence of which t-score produces which p-value after integrating the corresponding CDF.
Reporting p<0.05 is as meaningful/meaningless as reporting t-score > 1.64485 (Normal CDF 1 tail test) .
Instead of reporting the p<3524.1646*10^-300 one could just report t-score > 50 too.

This is a cultural issue of how the workflow has always been and it really needs revisiting.
As a side side note, perhaps since all that matters is if a line is present (correlation) what about proposing a less expensive method that looks for this relation in very loose terms? Since the result really doesn't matter beyond identifying a "signal" for further study, the whole GWAS could be revisited to still give 'meaningful' results by identifying slopes ina visual or similar manner. Just a thought.

A Frank




________________________________
From: genabel-devel-bounces at lists.r-forge.r-project.org [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of Frank, Alvaro Jesus [alvaro.frank at rwth-aachen.de]
Sent: Monday, June 23, 2014 4:02 PM
To: L.C. Karssen; genabel-devel at lists.r-forge.r-project.org
Subject: Re: [GenABEL-dev] P-values

That sounds cool. An (unnecessary?) word of caution though: I don't know
how much research you have done on this before starting, but make sure
to look at existing implementations of p-value calculation. Given that
in genetics p-values can become extremely small (e.g. 1e-100 or smaller)
numerical problems abound.
See also http://dx.doi.org/10.1016/j.csda.2008.11.028

Be sure to check out existing implementations (e.g. in Boost) on how
integration is being handled. Somewhere at the back of my mind the term
'incomplete Gamma function' comes to mind.

Such extreme values are directly related to the t-score, and it starts being a problem at around 20+.
For those cases I am set to use arbitrary precise functions as in: http://www.mpfr.org/mpfr-current/mpfr.html

Which leads me to the critic that if people only report p-values as their work, perhaps there is a cultural and scientific problem with the workflow and conceptual basis. But if they want their 3.5*10^-300 I will calculate it.

I would make the actual threshold a user-definable option, by the way.
There may be use cases where people want to have accurate p-values even
for non-significant cases.
I was thinking the same. From Local workflows here at Helmholtz I see that any insignificant value gets ignored completely unless it comes to doing Manhattan plots. So another setting would be to not calculate or store at all anything for non significant t-scores, which is helpfull during early discovery of significant signals. For Final calculations for plotting and reporting, the setting can then be changed to enable those. This effectivly removes the "Big data" problem, since 99% of the data is meaningless to determining whether the slope is significant.

In ProbABEL we are now implementing the calculation of p-values, but the
actual calculation is being left to the Boost libraries. Currently we
calculate the p-values irrespective of the value of the statistic. I'll
profile the code to see how much time this takes and whether it makes
sense to only calculate p-values for statistics with a value above a
certain threshold.

I was considering boost, if your profiling is good, perhaps boost is doing the threshold already. Let me know to see if I should switch over to boost too.

-A Frank
________________________________________
From: genabel-devel-bounces at lists.r-forge.r-project.org [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of L.C. Karssen [lennart at karssen.org]
Sent: Monday, June 23, 2014 3:51 PM
To: genabel-devel at lists.r-forge.r-project.org
Subject: Re: [GenABEL-dev] P-values

Hi Alvaro,

On 23-06-14 15:21, Frank, Alvaro Jesus wrote:
> Hi all,
>
> I have been trying to figure out an efficient way to calculate p-values,
> and it seems that I managed to come to an efficient compromise between
> speed an accuracy.

That sounds cool. An (unnecessary?) word of caution though: I don't know
how much research you have done on this before starting, but make sure
to look at existing implementations of p-value calculation. Given that
in genetics p-values can become extremely small (e.g. 1e-100 or smaller)
numerical problems abound.
See also http://dx.doi.org/10.1016/j.csda.2008.11.028

> 99% of the regressions will yield a non significant p
> value, that is to say p > 0.05 or even conservatively 0.1 . It is easy
> to know apriori if the p-value is significant, by looking at the t-score
> where it originates from. For any t-score < 1.28 the pvalue will not go
> below 0.1 for a t-distribution or normal distribution. In this cases
> (9X%) an aproximation with an error of 10^-(4~5) of the p-value is

Do you mean an absolute or relative error of 1e-4 here?

> enough. This calculation is efficient involving only a quadratic
> polynomial to be approximated. For possible significant p-values, with
> t-score > 1.28 a proper calculation of the p-value can be done.

I like the idea of setting a threshold for t, below which doing slow but
accurate calculations make no sense. If that gives a considerable speed
up, go for it!
I would make the actual threshold a user-definable option, by the way.
There may be use cases where people want to have accurate p-values even
for non-significant cases.

> Note
> that a p-value calculation involves approximating the integral of the
> distribution used, either t-students (n<1000?) or normal distribution.

Be sure to check out existing implementations (e.g. in Boost) on how
integration is being handled. Somewhere at the back of my mind the term
'incomplete Gamma function' comes to mind.

>
> How are the different genabel packages handling this at the moment? This
> is a speedup that can be applied to any p-value calculation.

The R-based packages will most likely use the functions provided by R,
e.g. pchisq() for calculating the p-value of a chi^2 statistic.
In ProbABEL we are now implementing the calculation of p-values, but the
actual calculation is being left to the Boost libraries. Currently we
calculate the p-values irrespective of the value of the statistic. I'll
profile the code to see how much time this takes and whether it makes
sense to only calculate p-values for statistics with a value above a
certain threshold.

>
> Plotted Error between 1-ncdf(x) and polynomial approx:
>
> http://www.wolframalpha.com/input/?i=y+%3D+%28%281%2F2+-+1%2F2*erf%28x%2Fsqrt%282%29%29%29+-+%281%2F2-%280.1*x*%284.4-x%29%29%29%29+from+0+to+1.28
>

ah, that leads me to believe the answer to my question above is that you
mean an absolute error of ~1e-4, correct?


Best,

Lennart.

>
> -Alvaro
>
>
> _______________________________________________
> genabel-devel mailing list
> genabel-devel at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
>

--
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140623/6c10f3b7/attachment.html>

From lennart at karssen.org  Mon Jun 23 17:38:28 2014
From: lennart at karssen.org (L.C. Karssen)
Date: Mon, 23 Jun 2014 17:38:28 +0200
Subject: [GenABEL-dev] P-values
In-Reply-To: <244CF001646FF74FB34F372310A332C57BBDF2@MBX5.rwth-ad.de>
References: <244CF001646FF74FB34F372310A332C57BBD8F@MBX5.rwth-ad.de>,
 <53A830D9.8020008@karssen.org>,
 <244CF001646FF74FB34F372310A332C57BBDD8@MBX5.rwth-ad.de>
 <244CF001646FF74FB34F372310A332C57BBDF2@MBX5.rwth-ad.de>
Message-ID: <53A849F4.5070903@karssen.org>

Hi Alvaro,

On 23-06-14 16:38, Frank, Alvaro Jesus wrote:
> Sorry for the p-value spam, but...

:-).

> 
> On Another note, it is important to note that a t-score is equivalent
> (in the 2 tailed case or a well defined head/tail) to any p-value since
> there is a 1-1 correspondence of which t-score produces which p-value
> after integrating the corresponding CDF.

I completely agree. However (and maybe superfluous), note that p-values
can also be calculated for statistics other than the t-statistic. In
that sense I can appreciate the fact that the p-value can be used to
'define' significance. Instead of remembering significance thresholds
for the t-distribution, chi^2 distribution, Z-distribution,
F-distribution and others, I can simply remember the p-value threshold.

That said, you are not the only one concerned about the simple reporting
of p-values or the taking of 0.05 as a strict threshold for
significance. See for example this recent feature in Nature:
http://dx.doi.org/10.1038/506150a

Another very good read is the paper "Why most published research
findings are false" by epidemiologist John Ioannidis, which puts the
p-value debate in a broader context by looking at reproducibility.

Back in 1998 Kenneth Rothman (a renowned epidemiologist) even
discouraged the use of p-values from the journal "Epidemiology", of
which he was the editor, see
http://journals.lww.com/epidem/Citation/1998/01000/That_Confounded_P_Value_.4.aspx.
It looks like he didn't succeed...

> Reporting p<0.05 is as meaningful/meaningless as reporting t-score >
> 1.64485 (Normal CDF 1 tail test) .
> Instead of reporting the p<3524.1646*10^-300 one could just report
> t-score > 50 too.

Yup. And numerically that would be more easy to do as well.

> 
> This is a cultural issue of how the workflow has always been and it
> really needs revisiting.
> As a side side note, perhaps since all that matters is if a line is
> present (correlation) what about proposing a less expensive method that
> looks for this relation in very loose terms? Since the result really
> doesn't matter beyond identifying a "signal" for further study, the
> whole GWAS could be revisited to still give 'meaningful' results by
> identifying slopes ina visual or similar manner. Just a thought.

Not only that, but remember that a p-value are the result of some form
of hypothesis testing. So it's not only about whether a certain
coefficient is different from zero (the usual hypothesis), but also
whether the alternative hypothesis makes sense in the first place.


While we're at it :-), here are few other things that should be changed
in the culture of the field:
- coefficients have physical units: these should be reported! So instead
of writing "beta is 3" one should add units (e.g. km/h, mmol/l per copy
of the A allele, kg/year, whatever).
- number of significant digits that are reported (and doing so
consistently throughout the paper).
- version numbers of software used for a given analysis should be reported.

A lot of these issues are addressed every once in a while, and until the
change in culture has taken place, the least thing you can do is keep
arguing the case(s) each time they come up (or when giving lectures, etc.).


Best,

Lennart.


> 
> A Frank
> 
> 
> 
> 
> ------------------------------------------------------------------------
> *From:* genabel-devel-bounces at lists.r-forge.r-project.org
> [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of Frank,
> Alvaro Jesus [alvaro.frank at rwth-aachen.de]
> *Sent:* Monday, June 23, 2014 4:02 PM
> *To:* L.C. Karssen; genabel-devel at lists.r-forge.r-project.org
> *Subject:* Re: [GenABEL-dev] P-values
> 
>     That sounds cool. An (unnecessary?) word of caution though: I don't know
>     how much research you have done on this before starting, but make sure
>     to look at existing implementations of p-value calculation. Given that
>     in genetics p-values can become extremely small (e.g. 1e-100 or smaller)
>     numerical problems abound.
>     See also http://dx.doi.org/10.1016/j.csda.2008.11.028
> 
>     Be sure to check out existing implementations (e.g. in Boost) on how
>     integration is being handled. Somewhere at the back of my mind the term
>     'incomplete Gamma function' comes to mind.
> 
> Such extreme values are directly related to the t-score, and it starts
> being a problem at around 20+.
> For those cases I am set to use arbitrary precise functions as in:
> http://www.mpfr.org/mpfr-current/mpfr.html
> 
> Which leads me to the critic that if people only report p-values as
> their work, perhaps there is a cultural and scientific problem with the
> workflow and conceptual basis. But if they want their 3.5*10^-300 I will
> calculate it.
> 
>     I would make the actual threshold a user-definable option, by the way.
>     There may be use cases where people want to have accurate p-values even
>     for non-significant cases.
> 
> I was thinking the same. From Local workflows here at Helmholtz I see
> that any insignificant value gets ignored completely unless it comes to
> doing Manhattan plots. So another setting would be to not calculate or
> store at all anything for non significant t-scores, which is helpfull
> during early discovery of significant signals. For Final calculations
> for plotting and reporting, the setting can then be changed to enable
> those. This effectivly removes the "Big data" problem, since 99% of the
> data is meaningless to determining whether the slope is significant.
> 
>     In ProbABEL we are now implementing the calculation of p-values, but the
>     actual calculation is being left to the Boost libraries. Currently we
>     calculate the p-values irrespective of the value of the statistic. I'll
>     profile the code to see how much time this takes and whether it makes
>     sense to only calculate p-values for statistics with a value above a
>     certain threshold.
> 
> I was considering boost, if your profiling is good, perhaps boost is
> doing the threshold already. Let me know to see if I should switch over
> to boost too.
> 
> -A Frank
> ________________________________________
> From: genabel-devel-bounces at lists.r-forge.r-project.org
> [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of L.C.
> Karssen [lennart at karssen.org]
> Sent: Monday, June 23, 2014 3:51 PM
> To: genabel-devel at lists.r-forge.r-project.org
> Subject: Re: [GenABEL-dev] P-values
> 
> Hi Alvaro,
> 
> On 23-06-14 15:21, Frank, Alvaro Jesus wrote:
>> Hi all,
>>
>> I have been trying to figure out an efficient way to calculate p-values,
>> and it seems that I managed to come to an efficient compromise between
>> speed an accuracy.
> 
> That sounds cool. An (unnecessary?) word of caution though: I don't know
> how much research you have done on this before starting, but make sure
> to look at existing implementations of p-value calculation. Given that
> in genetics p-values can become extremely small (e.g. 1e-100 or smaller)
> numerical problems abound.
> See also http://dx.doi.org/10.1016/j.csda.2008.11.028
> 
>> 99% of the regressions will yield a non significant p
>> value, that is to say p > 0.05 or even conservatively 0.1 . It is easy
>> to know apriori if the p-value is significant, by looking at the t-score
>> where it originates from. For any t-score < 1.28 the pvalue will not go
>> below 0.1 for a t-distribution or normal distribution. In this cases
>> (9X%) an aproximation with an error of 10^-(4~5) of the p-value is
> 
> Do you mean an absolute or relative error of 1e-4 here?
> 
>> enough. This calculation is efficient involving only a quadratic
>> polynomial to be approximated. For possible significant p-values, with
>> t-score > 1.28 a proper calculation of the p-value can be done.
> 
> I like the idea of setting a threshold for t, below which doing slow but
> accurate calculations make no sense. If that gives a considerable speed
> up, go for it!
> I would make the actual threshold a user-definable option, by the way.
> There may be use cases where people want to have accurate p-values even
> for non-significant cases.
> 
>> Note
>> that a p-value calculation involves approximating the integral of the
>> distribution used, either t-students (n<1000?) or normal distribution.
> 
> Be sure to check out existing implementations (e.g. in Boost) on how
> integration is being handled. Somewhere at the back of my mind the term
> 'incomplete Gamma function' comes to mind.
> 
>>
>> How are the different genabel packages handling this at the moment? This
>> is a speedup that can be applied to any p-value calculation.
> 
> The R-based packages will most likely use the functions provided by R,
> e.g. pchisq() for calculating the p-value of a chi^2 statistic.
> In ProbABEL we are now implementing the calculation of p-values, but the
> actual calculation is being left to the Boost libraries. Currently we
> calculate the p-values irrespective of the value of the statistic. I'll
> profile the code to see how much time this takes and whether it makes
> sense to only calculate p-values for statistics with a value above a
> certain threshold.
> 
>>
>> Plotted Error between 1-ncdf(x) and polynomial approx:
>>
>>
> http://www.wolframalpha.com/input/?i=y+%3D+%28%281%2F2+-+1%2F2*erf%28x%2Fsqrt%282%29%29%29+-+%281%2F2-%280.1*x*%284.4-x%29%29%29%29+from+0+to+1.28
>>
> 
> ah, that leads me to believe the answer to my question above is that you
> mean an absolute error of ~1e-4, correct?
> 
> 
> Best,
> 
> Lennart.
> 
>>
>> -Alvaro
>>
>>
>> _______________________________________________
>> genabel-devel mailing list
>> genabel-devel at lists.r-forge.r-project.org
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
>>
> 
> --
> *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
> L.C. Karssen
> Utrecht
> The Netherlands
> 
> lennart at karssen.org
> http://blog.karssen.org
> GPG key ID: A88F554A
> -*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-
> 

-- 
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 213 bytes
Desc: OpenPGP digital signature
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140623/7a9ef3eb/attachment-0001.sig>

From lennart at karssen.org  Mon Jun 23 18:39:35 2014
From: lennart at karssen.org (L.C. Karssen)
Date: Mon, 23 Jun 2014 18:39:35 +0200
Subject: [GenABEL-dev] P-values
In-Reply-To: <244CF001646FF74FB34F372310A332C57BBDD8@MBX5.rwth-ad.de>
References: <244CF001646FF74FB34F372310A332C57BBD8F@MBX5.rwth-ad.de>,
 <53A830D9.8020008@karssen.org>
 <244CF001646FF74FB34F372310A332C57BBDD8@MBX5.rwth-ad.de>
Message-ID: <53A85847.3050600@karssen.org>

Hi Alvaro,

On 23-06-14 16:02, Frank, Alvaro Jesus wrote:
>     That sounds cool. An (unnecessary?) word of caution though: I don't know
>     how much research you have done on this before starting, but make sure
>     to look at existing implementations of p-value calculation. Given that
>     in genetics p-values can become extremely small (e.g. 1e-100 or smaller)
>     numerical problems abound.
>     See also http://dx.doi.org/10.1016/j.csda.2008.11.028
> 
>     Be sure to check out existing implementations (e.g. in Boost) on how
>     integration is being handled. Somewhere at the back of my mind the term
>     'incomplete Gamma function' comes to mind.
> 
> Such extreme values are directly related to the t-score, and it starts
> being a problem at around 20+.
> For those cases I am set to use arbitrary precise functions as in:
> http://www.mpfr.org/mpfr-current/mpfr.html

Thanks for the link. I'm not really familiar with arbitrary precision
math, and this seems like a good place to start reading and playing.

> 
> Which leads me to the critic that if people only report p-values as
> their work, perhaps there is a cultural and scientific problem with the
> workflow and conceptual basis. But if they want their 3.5*10^-300 I will
> calculate it.

For now, that's the way life is. ;-)

> 
>     I would make the actual threshold a user-definable option, by the way.
>     There may be use cases where people want to have accurate p-values even
>     for non-significant cases.
> 
> I was thinking the same. From Local workflows here at Helmholtz I see
> that any insignificant value gets ignored completely unless it comes to
> doing Manhattan plots.

That's mostly my experience as well. However, when it comes to
meta-analysing data from different groups data (sometimes including
p-values) need to be present for all SNPs.

> So another setting would be to not calculate or
> store at all anything for non significant t-scores, which is helpfull
> during early discovery of significant signals. 

I think that such an option would be a good thing.


> For Final calculations
> for plotting and reporting, the setting can then be changed to enable
> those. This effectivly removes the "Big data" problem, since 99% of the
> data is meaningless to determining whether the slope is significant.

I agree.


> 
>     In ProbABEL we are now implementing the calculation of p-values, but the
>     actual calculation is being left to the Boost libraries. Currently we
>     calculate the p-values irrespective of the value of the statistic. I'll
>     profile the code to see how much time this takes and whether it makes
>     sense to only calculate p-values for statistics with a value above a
>     certain threshold.
> 
> I was considering boost, if your profiling is good, perhaps boost is
> doing the threshold already. Let me know to see if I should switch over
> to boost too.

I'm not familiar with MPFR, but from a maintainability point of view it
may be worth considering Boost since we're already using that in
ProbABEL. Moreover, since Boost contains many other algorithms and is
extensively used I also assume it is well-tested. Another point to
consider is that Boost is C++ whereas MPFR is C. Since most of the code
we write is C++ it makes sense to go for Boost (again, from a
maintainability point of view; from a performance PoV that may be
different).

As to the profiling: I'll keep you posted!


Best,

Lennart.


> 
> -A Frank
> ________________________________________
> From: genabel-devel-bounces at lists.r-forge.r-project.org
> [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of L.C.
> Karssen [lennart at karssen.org]
> Sent: Monday, June 23, 2014 3:51 PM
> To: genabel-devel at lists.r-forge.r-project.org
> Subject: Re: [GenABEL-dev] P-values
> 
> Hi Alvaro,
> 
> On 23-06-14 15:21, Frank, Alvaro Jesus wrote:
>> Hi all,
>>
>> I have been trying to figure out an efficient way to calculate p-values,
>> and it seems that I managed to come to an efficient compromise between
>> speed an accuracy.
> 
> That sounds cool. An (unnecessary?) word of caution though: I don't know
> how much research you have done on this before starting, but make sure
> to look at existing implementations of p-value calculation. Given that
> in genetics p-values can become extremely small (e.g. 1e-100 or smaller)
> numerical problems abound.
> See also http://dx.doi.org/10.1016/j.csda.2008.11.028
> 
>> 99% of the regressions will yield a non significant p
>> value, that is to say p > 0.05 or even conservatively 0.1 . It is easy
>> to know apriori if the p-value is significant, by looking at the t-score
>> where it originates from. For any t-score < 1.28 the pvalue will not go
>> below 0.1 for a t-distribution or normal distribution. In this cases
>> (9X%) an aproximation with an error of 10^-(4~5) of the p-value is
> 
> Do you mean an absolute or relative error of 1e-4 here?
> 
>> enough. This calculation is efficient involving only a quadratic
>> polynomial to be approximated. For possible significant p-values, with
>> t-score > 1.28 a proper calculation of the p-value can be done.
> 
> I like the idea of setting a threshold for t, below which doing slow but
> accurate calculations make no sense. If that gives a considerable speed
> up, go for it!
> I would make the actual threshold a user-definable option, by the way.
> There may be use cases where people want to have accurate p-values even
> for non-significant cases.
> 
>> Note
>> that a p-value calculation involves approximating the integral of the
>> distribution used, either t-students (n<1000?) or normal distribution.
> 
> Be sure to check out existing implementations (e.g. in Boost) on how
> integration is being handled. Somewhere at the back of my mind the term
> 'incomplete Gamma function' comes to mind.
> 
>>
>> How are the different genabel packages handling this at the moment? This
>> is a speedup that can be applied to any p-value calculation.
> 
> The R-based packages will most likely use the functions provided by R,
> e.g. pchisq() for calculating the p-value of a chi^2 statistic.
> In ProbABEL we are now implementing the calculation of p-values, but the
> actual calculation is being left to the Boost libraries. Currently we
> calculate the p-values irrespective of the value of the statistic. I'll
> profile the code to see how much time this takes and whether it makes
> sense to only calculate p-values for statistics with a value above a
> certain threshold.
> 
>>
>> Plotted Error between 1-ncdf(x) and polynomial approx:
>>
>>
> http://www.wolframalpha.com/input/?i=y+%3D+%28%281%2F2+-+1%2F2*erf%28x%2Fsqrt%282%29%29%29+-+%281%2F2-%280.1*x*%284.4-x%29%29%29%29+from+0+to+1.28
>>
> 
> ah, that leads me to believe the answer to my question above is that you
> mean an absolute error of ~1e-4, correct?
> 
> 
> Best,
> 
> Lennart.
> 
>>
>> -Alvaro
>>
>>
>> _______________________________________________
>> genabel-devel mailing list
>> genabel-devel at lists.r-forge.r-project.org
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
>>
> 
> --
> *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
> L.C. Karssen
> Utrecht
> The Netherlands
> 
> lennart at karssen.org
> http://blog.karssen.org
> GPG key ID: A88F554A
> -*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-
> 

-- 
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 213 bytes
Desc: OpenPGP digital signature
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140623/f6786d40/attachment.sig>

From alvaro.frank at rwth-aachen.de  Tue Jun 24 11:51:06 2014
From: alvaro.frank at rwth-aachen.de (Frank, Alvaro Jesus)
Date: Tue, 24 Jun 2014 09:51:06 +0000
Subject: [GenABEL-dev] automake boost
Message-ID: <244CF001646FF74FB34F372310A332C57BC165@MBX5.rwth-ad.de>

Hi Lennart,

I am, having trouble with automake recognizing boost root path.
I used AX_BOOST_BASE but the root path (/usr/local/) is still not being recognized. and the path $Boost_root is still not helping me, perhaps I am setting it wrong.
Do you also compile it all or keep the macros as they are?

Thanks!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140624/f094ae2f/attachment.html>

From lennart at karssen.org  Tue Jun 24 12:02:14 2014
From: lennart at karssen.org (L.C. Karssen)
Date: Tue, 24 Jun 2014 12:02:14 +0200
Subject: [GenABEL-dev] automake boost
In-Reply-To: <244CF001646FF74FB34F372310A332C57BC165@MBX5.rwth-ad.de>
References: <244CF001646FF74FB34F372310A332C57BC165@MBX5.rwth-ad.de>
Message-ID: <53A94CA6.7030100@karssen.org>

Hi Alvaro,


On 24-06-14 11:51, Frank, Alvaro Jesus wrote:
> Hi Lennart,
> 
> I am, having trouble with automake recognizing boost root path.
> I used AX_BOOST_BASE but the root path (/usr/local/) is still not being
> recognized. and the path $Boost_root is still not helping me, perhaps I
> am setting it wrong.
> Do you also compile it all or keep the macros as they are?

I use the Boost package provided by Ubuntu, so I haven't compiled them
myself. In that case the libraries are installed in /usr/include/boost/.

In ProbABEL's configure.ac I test for boost as follows (see the branch
at
https://r-forge.r-project.org/scm/viewvc.php/branches/ProbABEL-pvals/ProbABEL/?root=genabel):

----------------------------------------
if test "x$with_boost_math" != "xno"; then
    AC_MSG_NOTICE([building using the Boost Math library enabled])

    AC_ARG_WITH([boost-include-path],
        [AS_HELP_STRING([--with-boost-include-path],
          [location of the Boost headers, defaults to /usr/include/boost])],
        [CXXFLAGS+=" -I${withval}"
        CPPFLAGS+=" -I${withval}"],
        [CXXFLAGS+=' -I/usr/include/boost'
        CPPFLAGS+=' -I/usr/include/boost'])

    # Check for the Boost Math header files
    AC_CHECK_HEADERS([boost/math/distributions.hpp])

    if test x$ac_cv_header_boost_math_distributions_hpp = xno; then
      AC_MSG_ERROR([Could not find the Boost Math header files. Did \
you specify --with-boost-include-path correctly? Or use --without-boost \
to disable the calculation of p-values.])
    fi
else
   AC_MSG_NOTICE([not using the Boost Math libraries, so no p-values in
the \
output])
fi
AM_CONDITIONAL([WITH_BOOST_MATH], test "x$with_boost_math" != "xno")
----------------------------------------

This creates a variable WITH_BOOST_MATH that can be used in AutoMake's
Makefile.am files:
----------------------------------------
if WITH_BOOST_MATH
palinear_CXXFLAGS += -DWITH_BOOST_MATH
endif
----------------------------------------

And then in the .cpp files, for example:
----------------------------------------
#if WITH_BOOST_MATH
    std::vector<std::ostringstream *> pval;
#endif
----------------------------------------


Hope that helps.

Lennart.


> 
> Thanks!
> 
> 
> _______________________________________________
> genabel-devel mailing list
> genabel-devel at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
> 

-- 
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 213 bytes
Desc: OpenPGP digital signature
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140624/211f13c9/attachment.sig>

From alvaro.frank at rwth-aachen.de  Tue Jun 24 13:19:23 2014
From: alvaro.frank at rwth-aachen.de (Frank, Alvaro Jesus)
Date: Tue, 24 Jun 2014 11:19:23 +0000
Subject: [GenABEL-dev] automake boost
In-Reply-To: <53A94CA6.7030100@karssen.org>
References: <244CF001646FF74FB34F372310A332C57BC165@MBX5.rwth-ad.de>,
 <53A94CA6.7030100@karssen.org>
Message-ID: <244CF001646FF74FB34F372310A332C57BC18E@MBX5.rwth-ad.de>

Thanks for the info,

I tried my version (which fails to recognize boost at all) and yours (which configure doesnt recognize boost but I can still get the compiler to include the files) and I get the following error (lots of similar ones after this):

/usr/include/boost/iterator/iterator_facade.hpp:43:19: error: expected identifier before ?(? token
   template <class I, class V, class TC, class R, class D> class iterator_facade;

by just including once the following:

#include <boost/math/special_functions/erf.hpp>

Any Ideas?
________________________________________
From: genabel-devel-bounces at lists.r-forge.r-project.org [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of L.C. Karssen [lennart at karssen.org]
Sent: Tuesday, June 24, 2014 12:02 PM
To: genabel-devel at lists.r-forge.r-project.org
Subject: Re: [GenABEL-dev] automake boost

Hi Alvaro,


On 24-06-14 11:51, Frank, Alvaro Jesus wrote:
> Hi Lennart,
>
> I am, having trouble with automake recognizing boost root path.
> I used AX_BOOST_BASE but the root path (/usr/local/) is still not being
> recognized. and the path $Boost_root is still not helping me, perhaps I
> am setting it wrong.
> Do you also compile it all or keep the macros as they are?

I use the Boost package provided by Ubuntu, so I haven't compiled them
myself. In that case the libraries are installed in /usr/include/boost/.

In ProbABEL's configure.ac I test for boost as follows (see the branch
at
https://r-forge.r-project.org/scm/viewvc.php/branches/ProbABEL-pvals/ProbABEL/?root=genabel):

----------------------------------------
if test "x$with_boost_math" != "xno"; then
    AC_MSG_NOTICE([building using the Boost Math library enabled])

    AC_ARG_WITH([boost-include-path],
        [AS_HELP_STRING([--with-boost-include-path],
          [location of the Boost headers, defaults to /usr/include/boost])],
        [CXXFLAGS+=" -I${withval}"
        CPPFLAGS+=" -I${withval}"],
        [CXXFLAGS+=' -I/usr/include/boost'
        CPPFLAGS+=' -I/usr/include/boost'])

    # Check for the Boost Math header files
    AC_CHECK_HEADERS([boost/math/distributions.hpp])

    if test x$ac_cv_header_boost_math_distributions_hpp = xno; then
      AC_MSG_ERROR([Could not find the Boost Math header files. Did \
you specify --with-boost-include-path correctly? Or use --without-boost \
to disable the calculation of p-values.])
    fi
else
   AC_MSG_NOTICE([not using the Boost Math libraries, so no p-values in
the \
output])
fi
AM_CONDITIONAL([WITH_BOOST_MATH], test "x$with_boost_math" != "xno")
----------------------------------------

This creates a variable WITH_BOOST_MATH that can be used in AutoMake's
Makefile.am files:
----------------------------------------
if WITH_BOOST_MATH
palinear_CXXFLAGS += -DWITH_BOOST_MATH
endif
----------------------------------------

And then in the .cpp files, for example:
----------------------------------------
#if WITH_BOOST_MATH
    std::vector<std::ostringstream *> pval;
#endif
----------------------------------------


Hope that helps.

Lennart.


>
> Thanks!
>
>
> _______________________________________________
> genabel-devel mailing list
> genabel-devel at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
>

--
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-


From alvaro.frank at rwth-aachen.de  Tue Jun 24 14:11:24 2014
From: alvaro.frank at rwth-aachen.de (Frank, Alvaro Jesus)
Date: Tue, 24 Jun 2014 12:11:24 +0000
Subject: [GenABEL-dev] automake boost
In-Reply-To: <244CF001646FF74FB34F372310A332C57BC18E@MBX5.rwth-ad.de>
References: <244CF001646FF74FB34F372310A332C57BC165@MBX5.rwth-ad.de>,
 <53A94CA6.7030100@karssen.org>,
 <244CF001646FF74FB34F372310A332C57BC18E@MBX5.rwth-ad.de>
Message-ID: <244CF001646FF74FB34F372310A332C57BC1AB@MBX5.rwth-ad.de>

I am truly stuck with this. The library installs correctly but including it just breaks the compile.


________________________________________
From: genabel-devel-bounces at lists.r-forge.r-project.org [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of Frank, Alvaro Jesus [alvaro.frank at rwth-aachen.de]
Sent: Tuesday, June 24, 2014 1:19 PM
To: L.C. Karssen; genabel-devel at lists.r-forge.r-project.org
Subject: Re: [GenABEL-dev] automake boost

Thanks for the info,

I tried my version (which fails to recognize boost at all) and yours (which configure doesnt recognize boost but I can still get the compiler to include the files) and I get the following error (lots of similar ones after this):

/usr/include/boost/iterator/iterator_facade.hpp:43:19: error: expected identifier before ?(? token
   template <class I, class V, class TC, class R, class D> class iterator_facade;

by just including once the following:

#include <boost/math/special_functions/erf.hpp>

Any Ideas?
________________________________________
From: genabel-devel-bounces at lists.r-forge.r-project.org [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of L.C. Karssen [lennart at karssen.org]
Sent: Tuesday, June 24, 2014 12:02 PM
To: genabel-devel at lists.r-forge.r-project.org
Subject: Re: [GenABEL-dev] automake boost

Hi Alvaro,


On 24-06-14 11:51, Frank, Alvaro Jesus wrote:
> Hi Lennart,
>
> I am, having trouble with automake recognizing boost root path.
> I used AX_BOOST_BASE but the root path (/usr/local/) is still not being
> recognized. and the path $Boost_root is still not helping me, perhaps I
> am setting it wrong.
> Do you also compile it all or keep the macros as they are?

I use the Boost package provided by Ubuntu, so I haven't compiled them
myself. In that case the libraries are installed in /usr/include/boost/.

In ProbABEL's configure.ac I test for boost as follows (see the branch
at
https://r-forge.r-project.org/scm/viewvc.php/branches/ProbABEL-pvals/ProbABEL/?root=genabel):

----------------------------------------
if test "x$with_boost_math" != "xno"; then
    AC_MSG_NOTICE([building using the Boost Math library enabled])

    AC_ARG_WITH([boost-include-path],
        [AS_HELP_STRING([--with-boost-include-path],
          [location of the Boost headers, defaults to /usr/include/boost])],
        [CXXFLAGS+=" -I${withval}"
        CPPFLAGS+=" -I${withval}"],
        [CXXFLAGS+=' -I/usr/include/boost'
        CPPFLAGS+=' -I/usr/include/boost'])

    # Check for the Boost Math header files
    AC_CHECK_HEADERS([boost/math/distributions.hpp])

    if test x$ac_cv_header_boost_math_distributions_hpp = xno; then
      AC_MSG_ERROR([Could not find the Boost Math header files. Did \
you specify --with-boost-include-path correctly? Or use --without-boost \
to disable the calculation of p-values.])
    fi
else
   AC_MSG_NOTICE([not using the Boost Math libraries, so no p-values in
the \
output])
fi
AM_CONDITIONAL([WITH_BOOST_MATH], test "x$with_boost_math" != "xno")
----------------------------------------

This creates a variable WITH_BOOST_MATH that can be used in AutoMake's
Makefile.am files:
----------------------------------------
if WITH_BOOST_MATH
palinear_CXXFLAGS += -DWITH_BOOST_MATH
endif
----------------------------------------

And then in the .cpp files, for example:
----------------------------------------
#if WITH_BOOST_MATH
    std::vector<std::ostringstream *> pval;
#endif
----------------------------------------


Hope that helps.

Lennart.


>
> Thanks!
>
>
> _______________________________________________
> genabel-devel mailing list
> genabel-devel at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
>

--
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

_______________________________________________
genabel-devel mailing list
genabel-devel at lists.r-forge.r-project.org
https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel

From lennart at karssen.org  Tue Jun 24 14:21:57 2014
From: lennart at karssen.org (L.C. Karssen)
Date: Tue, 24 Jun 2014 14:21:57 +0200
Subject: [GenABEL-dev] automake boost
In-Reply-To: <244CF001646FF74FB34F372310A332C57BC18E@MBX5.rwth-ad.de>
References: <244CF001646FF74FB34F372310A332C57BC165@MBX5.rwth-ad.de>,
 <53A94CA6.7030100@karssen.org>
 <244CF001646FF74FB34F372310A332C57BC18E@MBX5.rwth-ad.de>
Message-ID: <53A96D65.9080604@karssen.org>

Hi Alvaro,

So if I understand you correctly this is the situation:
- When using my version autoconf doesn't find boost. Correct?
- But nevertheless if you try to include the erf.hpp file (and manually
(?) instruct the compiler where to find the header files, you get the
error below.

How do you compile your code? Manually involing g++ or using a Makefile
generated by automake?

Could you provide me with a minimum working example (actually a minimum
breaking example) of configure.ac, Makefile.am and main.cpp to test?


Lennart.

On 24-06-14 13:19, Frank, Alvaro Jesus wrote:
> Thanks for the info,
> 
> I tried my version (which fails to recognize boost at all) and yours (which configure doesnt recognize boost but I can still get the compiler to include the files) and I get the following error (lots of similar ones after this):
> 
> /usr/include/boost/iterator/iterator_facade.hpp:43:19: error: expected identifier before ?(? token
>    template <class I, class V, class TC, class R, class D> class iterator_facade;
> 
> by just including once the following:
> 
> #include <boost/math/special_functions/erf.hpp>
> 
> Any Ideas?
> ________________________________________
> From: genabel-devel-bounces at lists.r-forge.r-project.org [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of L.C. Karssen [lennart at karssen.org]
> Sent: Tuesday, June 24, 2014 12:02 PM
> To: genabel-devel at lists.r-forge.r-project.org
> Subject: Re: [GenABEL-dev] automake boost
> 
> Hi Alvaro,
> 
> 
> On 24-06-14 11:51, Frank, Alvaro Jesus wrote:
>> Hi Lennart,
>>
>> I am, having trouble with automake recognizing boost root path.
>> I used AX_BOOST_BASE but the root path (/usr/local/) is still not being
>> recognized. and the path $Boost_root is still not helping me, perhaps I
>> am setting it wrong.
>> Do you also compile it all or keep the macros as they are?
> 
> I use the Boost package provided by Ubuntu, so I haven't compiled them
> myself. In that case the libraries are installed in /usr/include/boost/.
> 
> In ProbABEL's configure.ac I test for boost as follows (see the branch
> at
> https://r-forge.r-project.org/scm/viewvc.php/branches/ProbABEL-pvals/ProbABEL/?root=genabel):
> 
> ----------------------------------------
> if test "x$with_boost_math" != "xno"; then
>     AC_MSG_NOTICE([building using the Boost Math library enabled])
> 
>     AC_ARG_WITH([boost-include-path],
>         [AS_HELP_STRING([--with-boost-include-path],
>           [location of the Boost headers, defaults to /usr/include/boost])],
>         [CXXFLAGS+=" -I${withval}"
>         CPPFLAGS+=" -I${withval}"],
>         [CXXFLAGS+=' -I/usr/include/boost'
>         CPPFLAGS+=' -I/usr/include/boost'])
> 
>     # Check for the Boost Math header files
>     AC_CHECK_HEADERS([boost/math/distributions.hpp])
> 
>     if test x$ac_cv_header_boost_math_distributions_hpp = xno; then
>       AC_MSG_ERROR([Could not find the Boost Math header files. Did \
> you specify --with-boost-include-path correctly? Or use --without-boost \
> to disable the calculation of p-values.])
>     fi
> else
>    AC_MSG_NOTICE([not using the Boost Math libraries, so no p-values in
> the \
> output])
> fi
> AM_CONDITIONAL([WITH_BOOST_MATH], test "x$with_boost_math" != "xno")
> ----------------------------------------
> 
> This creates a variable WITH_BOOST_MATH that can be used in AutoMake's
> Makefile.am files:
> ----------------------------------------
> if WITH_BOOST_MATH
> palinear_CXXFLAGS += -DWITH_BOOST_MATH
> endif
> ----------------------------------------
> 
> And then in the .cpp files, for example:
> ----------------------------------------
> #if WITH_BOOST_MATH
>     std::vector<std::ostringstream *> pval;
> #endif
> ----------------------------------------
> 
> 
> Hope that helps.
> 
> Lennart.
> 
> 
>>
>> Thanks!
>>
>>
>> _______________________________________________
>> genabel-devel mailing list
>> genabel-devel at lists.r-forge.r-project.org
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
>>
> 
> --
> *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
> L.C. Karssen
> Utrecht
> The Netherlands
> 
> lennart at karssen.org
> http://blog.karssen.org
> GPG key ID: A88F554A
> -*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-
> 

-- 
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 213 bytes
Desc: OpenPGP digital signature
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140624/b5ca5a38/attachment.sig>

From alvaro.frank at rwth-aachen.de  Tue Jun 24 14:25:28 2014
From: alvaro.frank at rwth-aachen.de (Frank, Alvaro Jesus)
Date: Tue, 24 Jun 2014 12:25:28 +0000
Subject: [GenABEL-dev] automake boost
In-Reply-To: <53A96D65.9080604@karssen.org>
References: <244CF001646FF74FB34F372310A332C57BC165@MBX5.rwth-ad.de>,
 <53A94CA6.7030100@karssen.org>
 <244CF001646FF74FB34F372310A332C57BC18E@MBX5.rwth-ad.de>,
 <53A96D65.9080604@karssen.org>
Message-ID: <244CF001646FF74FB34F372310A332C57BC1BF@MBX5.rwth-ad.de>

Using any working copy of boost, including the file #include <boost/..../anything.hpp>
will lead to a failed compilation IF the file was NOT included as the first file in the main.cpp
which is ridiculous.

I got it working right now. It could be some kind of redefinition of something relevant only to this code.
________________________________________
From: L.C. Karssen [lennart at karssen.org]
Sent: Tuesday, June 24, 2014 2:21 PM
To: Frank, Alvaro Jesus; genabel-devel at lists.r-forge.r-project.org
Subject: Re: [GenABEL-dev] automake boost

Hi Alvaro,

So if I understand you correctly this is the situation:
- When using my version autoconf doesn't find boost. Correct?
- But nevertheless if you try to include the erf.hpp file (and manually
(?) instruct the compiler where to find the header files, you get the
error below.

How do you compile your code? Manually involing g++ or using a Makefile
generated by automake?

Could you provide me with a minimum working example (actually a minimum
breaking example) of configure.ac, Makefile.am and main.cpp to test?


Lennart.

On 24-06-14 13:19, Frank, Alvaro Jesus wrote:
> Thanks for the info,
>
> I tried my version (which fails to recognize boost at all) and yours (which configure doesnt recognize boost but I can still get the compiler to include the files) and I get the following error (lots of similar ones after this):
>
> /usr/include/boost/iterator/iterator_facade.hpp:43:19: error: expected identifier before ?(? token
>    template <class I, class V, class TC, class R, class D> class iterator_facade;
>
> by just including once the following:
>
> #include <boost/math/special_functions/erf.hpp>
>
> Any Ideas?
> ________________________________________
> From: genabel-devel-bounces at lists.r-forge.r-project.org [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of L.C. Karssen [lennart at karssen.org]
> Sent: Tuesday, June 24, 2014 12:02 PM
> To: genabel-devel at lists.r-forge.r-project.org
> Subject: Re: [GenABEL-dev] automake boost
>
> Hi Alvaro,
>
>
> On 24-06-14 11:51, Frank, Alvaro Jesus wrote:
>> Hi Lennart,
>>
>> I am, having trouble with automake recognizing boost root path.
>> I used AX_BOOST_BASE but the root path (/usr/local/) is still not being
>> recognized. and the path $Boost_root is still not helping me, perhaps I
>> am setting it wrong.
>> Do you also compile it all or keep the macros as they are?
>
> I use the Boost package provided by Ubuntu, so I haven't compiled them
> myself. In that case the libraries are installed in /usr/include/boost/.
>
> In ProbABEL's configure.ac I test for boost as follows (see the branch
> at
> https://r-forge.r-project.org/scm/viewvc.php/branches/ProbABEL-pvals/ProbABEL/?root=genabel):
>
> ----------------------------------------
> if test "x$with_boost_math" != "xno"; then
>     AC_MSG_NOTICE([building using the Boost Math library enabled])
>
>     AC_ARG_WITH([boost-include-path],
>         [AS_HELP_STRING([--with-boost-include-path],
>           [location of the Boost headers, defaults to /usr/include/boost])],
>         [CXXFLAGS+=" -I${withval}"
>         CPPFLAGS+=" -I${withval}"],
>         [CXXFLAGS+=' -I/usr/include/boost'
>         CPPFLAGS+=' -I/usr/include/boost'])
>
>     # Check for the Boost Math header files
>     AC_CHECK_HEADERS([boost/math/distributions.hpp])
>
>     if test x$ac_cv_header_boost_math_distributions_hpp = xno; then
>       AC_MSG_ERROR([Could not find the Boost Math header files. Did \
> you specify --with-boost-include-path correctly? Or use --without-boost \
> to disable the calculation of p-values.])
>     fi
> else
>    AC_MSG_NOTICE([not using the Boost Math libraries, so no p-values in
> the \
> output])
> fi
> AM_CONDITIONAL([WITH_BOOST_MATH], test "x$with_boost_math" != "xno")
> ----------------------------------------
>
> This creates a variable WITH_BOOST_MATH that can be used in AutoMake's
> Makefile.am files:
> ----------------------------------------
> if WITH_BOOST_MATH
> palinear_CXXFLAGS += -DWITH_BOOST_MATH
> endif
> ----------------------------------------
>
> And then in the .cpp files, for example:
> ----------------------------------------
> #if WITH_BOOST_MATH
>     std::vector<std::ostringstream *> pval;
> #endif
> ----------------------------------------
>
>
> Hope that helps.
>
> Lennart.
>
>
>>
>> Thanks!
>>
>>
>> _______________________________________________
>> genabel-devel mailing list
>> genabel-devel at lists.r-forge.r-project.org
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
>>
>
> --
> *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
> L.C. Karssen
> Utrecht
> The Netherlands
>
> lennart at karssen.org
> http://blog.karssen.org
> GPG key ID: A88F554A
> -*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-
>

--
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-


From lennart at karssen.org  Tue Jun 24 14:27:22 2014
From: lennart at karssen.org (L.C. Karssen)
Date: Tue, 24 Jun 2014 14:27:22 +0200
Subject: [GenABEL-dev] automake boost
In-Reply-To: <53A96D65.9080604@karssen.org>
References: <244CF001646FF74FB34F372310A332C57BC165@MBX5.rwth-ad.de>,
 <53A94CA6.7030100@karssen.org>
 <244CF001646FF74FB34F372310A332C57BC18E@MBX5.rwth-ad.de>
 <53A96D65.9080604@karssen.org>
Message-ID: <53A96EAA.8040806@karssen.org>

Hi Alvaro,

A quick follow-up: I dug up the attached file with which I made my first
Boost steps (exploring how to compute p-values for ProbABEL).

If you run
  make p-from-chi2
in the directory that contains the .cpp file it should compile if Boost
is in a standard location (no Makefile needed because make will use its
built-in rules). Otherwise
  g++ p-from-chi2.cpp   -o p-from-chi2
with the appropriate -I lines for the Boost include files should work.

Lennart.

On 24-06-14 14:21, L.C. Karssen wrote:
> Hi Alvaro,
> 
> So if I understand you correctly this is the situation:
> - When using my version autoconf doesn't find boost. Correct?
> - But nevertheless if you try to include the erf.hpp file (and manually
> (?) instruct the compiler where to find the header files, you get the
> error below.
> 
> How do you compile your code? Manually involing g++ or using a Makefile
> generated by automake?
> 
> Could you provide me with a minimum working example (actually a minimum
> breaking example) of configure.ac, Makefile.am and main.cpp to test?
> 
> 
> Lennart.
> 
> On 24-06-14 13:19, Frank, Alvaro Jesus wrote:
>> Thanks for the info,
>>
>> I tried my version (which fails to recognize boost at all) and yours (which configure doesnt recognize boost but I can still get the compiler to include the files) and I get the following error (lots of similar ones after this):
>>
>> /usr/include/boost/iterator/iterator_facade.hpp:43:19: error: expected identifier before ?(? token
>>    template <class I, class V, class TC, class R, class D> class iterator_facade;
>>
>> by just including once the following:
>>
>> #include <boost/math/special_functions/erf.hpp>
>>
>> Any Ideas?
>> ________________________________________
>> From: genabel-devel-bounces at lists.r-forge.r-project.org [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of L.C. Karssen [lennart at karssen.org]
>> Sent: Tuesday, June 24, 2014 12:02 PM
>> To: genabel-devel at lists.r-forge.r-project.org
>> Subject: Re: [GenABEL-dev] automake boost
>>
>> Hi Alvaro,
>>
>>
>> On 24-06-14 11:51, Frank, Alvaro Jesus wrote:
>>> Hi Lennart,
>>>
>>> I am, having trouble with automake recognizing boost root path.
>>> I used AX_BOOST_BASE but the root path (/usr/local/) is still not being
>>> recognized. and the path $Boost_root is still not helping me, perhaps I
>>> am setting it wrong.
>>> Do you also compile it all or keep the macros as they are?
>>
>> I use the Boost package provided by Ubuntu, so I haven't compiled them
>> myself. In that case the libraries are installed in /usr/include/boost/.
>>
>> In ProbABEL's configure.ac I test for boost as follows (see the branch
>> at
>> https://r-forge.r-project.org/scm/viewvc.php/branches/ProbABEL-pvals/ProbABEL/?root=genabel):
>>
>> ----------------------------------------
>> if test "x$with_boost_math" != "xno"; then
>>     AC_MSG_NOTICE([building using the Boost Math library enabled])
>>
>>     AC_ARG_WITH([boost-include-path],
>>         [AS_HELP_STRING([--with-boost-include-path],
>>           [location of the Boost headers, defaults to /usr/include/boost])],
>>         [CXXFLAGS+=" -I${withval}"
>>         CPPFLAGS+=" -I${withval}"],
>>         [CXXFLAGS+=' -I/usr/include/boost'
>>         CPPFLAGS+=' -I/usr/include/boost'])
>>
>>     # Check for the Boost Math header files
>>     AC_CHECK_HEADERS([boost/math/distributions.hpp])
>>
>>     if test x$ac_cv_header_boost_math_distributions_hpp = xno; then
>>       AC_MSG_ERROR([Could not find the Boost Math header files. Did \
>> you specify --with-boost-include-path correctly? Or use --without-boost \
>> to disable the calculation of p-values.])
>>     fi
>> else
>>    AC_MSG_NOTICE([not using the Boost Math libraries, so no p-values in
>> the \
>> output])
>> fi
>> AM_CONDITIONAL([WITH_BOOST_MATH], test "x$with_boost_math" != "xno")
>> ----------------------------------------
>>
>> This creates a variable WITH_BOOST_MATH that can be used in AutoMake's
>> Makefile.am files:
>> ----------------------------------------
>> if WITH_BOOST_MATH
>> palinear_CXXFLAGS += -DWITH_BOOST_MATH
>> endif
>> ----------------------------------------
>>
>> And then in the .cpp files, for example:
>> ----------------------------------------
>> #if WITH_BOOST_MATH
>>     std::vector<std::ostringstream *> pval;
>> #endif
>> ----------------------------------------
>>
>>
>> Hope that helps.
>>
>> Lennart.
>>
>>
>>>
>>> Thanks!
>>>
>>>
>>> _______________________________________________
>>> genabel-devel mailing list
>>> genabel-devel at lists.r-forge.r-project.org
>>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
>>>
>>
>> --
>> *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
>> L.C. Karssen
>> Utrecht
>> The Netherlands
>>
>> lennart at karssen.org
>> http://blog.karssen.org
>> GPG key ID: A88F554A
>> -*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-
>>
> 
> 
> 
> _______________________________________________
> genabel-devel mailing list
> genabel-devel at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
> 

-- 
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-
-------------- next part --------------
A non-text attachment was scrubbed...
Name: p-from-chi2.cpp
Type: text/x-c++src
Size: 1186 bytes
Desc: not available
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140624/90071cdb/attachment.cpp>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 213 bytes
Desc: OpenPGP digital signature
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140624/90071cdb/attachment.sig>

From lennart at karssen.org  Tue Jun 24 14:28:40 2014
From: lennart at karssen.org (L.C. Karssen)
Date: Tue, 24 Jun 2014 14:28:40 +0200
Subject: [GenABEL-dev] automake boost
In-Reply-To: <244CF001646FF74FB34F372310A332C57BC1BF@MBX5.rwth-ad.de>
References: <244CF001646FF74FB34F372310A332C57BC165@MBX5.rwth-ad.de>,
 <53A94CA6.7030100@karssen.org>
 <244CF001646FF74FB34F372310A332C57BC18E@MBX5.rwth-ad.de>,
 <53A96D65.9080604@karssen.org>
 <244CF001646FF74FB34F372310A332C57BC1BF@MBX5.rwth-ad.de>
Message-ID: <53A96EF8.4040606@karssen.org>

On 24-06-14 14:25, Frank, Alvaro Jesus wrote:
> Using any working copy of boost, including the file #include <boost/..../anything.hpp>
> will lead to a failed compilation IF the file was NOT included as the first file in the main.cpp
> which is ridiculous.

I agree, that shouldn't happen. Weird!

> 
> I got it working right now. It could be some kind of redefinition of something relevant only to this code.


Good luck!


Lennart.

> ________________________________________
> From: L.C. Karssen [lennart at karssen.org]
> Sent: Tuesday, June 24, 2014 2:21 PM
> To: Frank, Alvaro Jesus; genabel-devel at lists.r-forge.r-project.org
> Subject: Re: [GenABEL-dev] automake boost
> 
> Hi Alvaro,
> 
> So if I understand you correctly this is the situation:
> - When using my version autoconf doesn't find boost. Correct?
> - But nevertheless if you try to include the erf.hpp file (and manually
> (?) instruct the compiler where to find the header files, you get the
> error below.
> 
> How do you compile your code? Manually involing g++ or using a Makefile
> generated by automake?
> 
> Could you provide me with a minimum working example (actually a minimum
> breaking example) of configure.ac, Makefile.am and main.cpp to test?
> 
> 
> Lennart.
> 
> On 24-06-14 13:19, Frank, Alvaro Jesus wrote:
>> Thanks for the info,
>>
>> I tried my version (which fails to recognize boost at all) and yours (which configure doesnt recognize boost but I can still get the compiler to include the files) and I get the following error (lots of similar ones after this):
>>
>> /usr/include/boost/iterator/iterator_facade.hpp:43:19: error: expected identifier before ?(? token
>>    template <class I, class V, class TC, class R, class D> class iterator_facade;
>>
>> by just including once the following:
>>
>> #include <boost/math/special_functions/erf.hpp>
>>
>> Any Ideas?
>> ________________________________________
>> From: genabel-devel-bounces at lists.r-forge.r-project.org [genabel-devel-bounces at lists.r-forge.r-project.org] on behalf of L.C. Karssen [lennart at karssen.org]
>> Sent: Tuesday, June 24, 2014 12:02 PM
>> To: genabel-devel at lists.r-forge.r-project.org
>> Subject: Re: [GenABEL-dev] automake boost
>>
>> Hi Alvaro,
>>
>>
>> On 24-06-14 11:51, Frank, Alvaro Jesus wrote:
>>> Hi Lennart,
>>>
>>> I am, having trouble with automake recognizing boost root path.
>>> I used AX_BOOST_BASE but the root path (/usr/local/) is still not being
>>> recognized. and the path $Boost_root is still not helping me, perhaps I
>>> am setting it wrong.
>>> Do you also compile it all or keep the macros as they are?
>>
>> I use the Boost package provided by Ubuntu, so I haven't compiled them
>> myself. In that case the libraries are installed in /usr/include/boost/.
>>
>> In ProbABEL's configure.ac I test for boost as follows (see the branch
>> at
>> https://r-forge.r-project.org/scm/viewvc.php/branches/ProbABEL-pvals/ProbABEL/?root=genabel):
>>
>> ----------------------------------------
>> if test "x$with_boost_math" != "xno"; then
>>     AC_MSG_NOTICE([building using the Boost Math library enabled])
>>
>>     AC_ARG_WITH([boost-include-path],
>>         [AS_HELP_STRING([--with-boost-include-path],
>>           [location of the Boost headers, defaults to /usr/include/boost])],
>>         [CXXFLAGS+=" -I${withval}"
>>         CPPFLAGS+=" -I${withval}"],
>>         [CXXFLAGS+=' -I/usr/include/boost'
>>         CPPFLAGS+=' -I/usr/include/boost'])
>>
>>     # Check for the Boost Math header files
>>     AC_CHECK_HEADERS([boost/math/distributions.hpp])
>>
>>     if test x$ac_cv_header_boost_math_distributions_hpp = xno; then
>>       AC_MSG_ERROR([Could not find the Boost Math header files. Did \
>> you specify --with-boost-include-path correctly? Or use --without-boost \
>> to disable the calculation of p-values.])
>>     fi
>> else
>>    AC_MSG_NOTICE([not using the Boost Math libraries, so no p-values in
>> the \
>> output])
>> fi
>> AM_CONDITIONAL([WITH_BOOST_MATH], test "x$with_boost_math" != "xno")
>> ----------------------------------------
>>
>> This creates a variable WITH_BOOST_MATH that can be used in AutoMake's
>> Makefile.am files:
>> ----------------------------------------
>> if WITH_BOOST_MATH
>> palinear_CXXFLAGS += -DWITH_BOOST_MATH
>> endif
>> ----------------------------------------
>>
>> And then in the .cpp files, for example:
>> ----------------------------------------
>> #if WITH_BOOST_MATH
>>     std::vector<std::ostringstream *> pval;
>> #endif
>> ----------------------------------------
>>
>>
>> Hope that helps.
>>
>> Lennart.
>>
>>
>>>
>>> Thanks!
>>>
>>>
>>> _______________________________________________
>>> genabel-devel mailing list
>>> genabel-devel at lists.r-forge.r-project.org
>>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/genabel-devel
>>>
>>
>> --
>> *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
>> L.C. Karssen
>> Utrecht
>> The Netherlands
>>
>> lennart at karssen.org
>> http://blog.karssen.org
>> GPG key ID: A88F554A
>> -*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-
>>
> 
> --
> *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
> L.C. Karssen
> Utrecht
> The Netherlands
> 
> lennart at karssen.org
> http://blog.karssen.org
> GPG key ID: A88F554A
> -*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-
> 

-- 
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
L.C. Karssen
Utrecht
The Netherlands

lennart at karssen.org
http://blog.karssen.org
GPG key ID: A88F554A
-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 213 bytes
Desc: OpenPGP digital signature
URL: <http://lists.r-forge.r-project.org/pipermail/genabel-devel/attachments/20140624/990109ab/attachment.sig>

From yurii.aulchenko at gmail.com  Fri Jun 27 14:29:11 2014
From: yurii.aulchenko at gmail.com (Yurii Aulchenko)
Date: Fri, 27 Jun 2014 14:29:11 +0200
Subject: [GenABEL-dev] Missing file for unit test in GenABEL
In-Reply-To: <20140626134702.GA19423@an3as.eu>
References: <20140620084443.GA17492@an3as.eu> <20140626134702.GA19423@an3as.eu>
Message-ID: <1145061290386285028@unknownmsgid>

Dear Andreas,

Sorry it delay with answer.

We specifically remove unit tests before submitting to cran. Our dev
repo does contain a version with unit tests, also tags for the
versions.

So one option would be to use our code repo instead of what is on
cran. What do you think?

Yurii


----------------------
Yurii Aulchenko
(sent from mobile device)

> On Jun 26, 2014, at 15:47, Andreas Tille <andreas at an3as.eu> wrote:
>
> Hi Yurii,
>
>> On Fri, Jun 20, 2014 at 10:44:43AM +0200, Andreas Tille wrote:
>> Hi Yurii,
>>
>> I'm trying to update the Debian package of GenABEL.  Since some time
>> there is an effort to automatically run unit tests of software if
>> available.  Since I noticed that GenABEL comes with unit tests I
>> tried
>>
>> $ make test
>> export RCMDCHECK=FALSE;\
>>        cd ../../tests;\
>>        R --vanilla --slave < doRUnit.R
>> /bin/sh: 2: cd: can't cd to ../../tests
>> /bin/sh: 3: cannot open doRUnit.R: No such file
>> make: *** [test] Error 2
>>
>>
>> As you can see the file doRUnit.R is missing.  It would be great if you
>> could include this file into the source diustribution to make sure we
>> can reproduce your exact test procedure in the Debian package.
>
> I cloned https://github.com/cran/GenABEL.git and had a look into the
> code.  I realised that the last tag containing the file doRUnit.R was
> 1.6-7.  I wonder how you are doing unit testing in the current version.
>
> Thanks for any enlightenment
>
>      Andreas.
>
> --
> http://fam-tille.de

From andreas at an3as.eu  Fri Jun 27 14:34:55 2014
From: andreas at an3as.eu (Andreas Tille)
Date: Fri, 27 Jun 2014 14:34:55 +0200
Subject: [GenABEL-dev] Missing file for unit test in GenABEL
In-Reply-To: <1145061290386285028@unknownmsgid>
References: <20140620084443.GA17492@an3as.eu> <20140626134702.GA19423@an3as.eu>
 <1145061290386285028@unknownmsgid>
Message-ID: <20140627123455.GC17339@an3as.eu>

Hi Yurii,

On Fri, Jun 27, 2014 at 02:29:11PM +0200, Yurii Aulchenko wrote:
> Dear Andreas,
> 
> Sorry it delay with answer.

No problem. :-)

> We specifically remove unit tests before submitting to cran. Our dev
> repo does contain a version with unit tests, also tags for the
> versions.

Do you have any reasons to remove the tests from the tarball?  I could
imagine that larger chunks of data might be used but I assumed that for
this very purpose GenABEL.data exists ... and so I did package it for
Debian as well.
 
> So one option would be to use our code repo instead of what is on
> cran. What do you think?

While this would possible in prinziple I personally really prefer a
downloadable source tarball including testing features if any possible.
However, if you might have strong reasons I would find ways to drain
the stuff needed from the repository.
 
Kind regards

      Andreas.
 
> ----------------------
> Yurii Aulchenko
> (sent from mobile device)
> 
> > On Jun 26, 2014, at 15:47, Andreas Tille <andreas at an3as.eu> wrote:
> >
> > Hi Yurii,
> >
> >> On Fri, Jun 20, 2014 at 10:44:43AM +0200, Andreas Tille wrote:
> >> Hi Yurii,
> >>
> >> I'm trying to update the Debian package of GenABEL.  Since some time
> >> there is an effort to automatically run unit tests of software if
> >> available.  Since I noticed that GenABEL comes with unit tests I
> >> tried
> >>
> >> $ make test
> >> export RCMDCHECK=FALSE;\
> >>        cd ../../tests;\
> >>        R --vanilla --slave < doRUnit.R
> >> /bin/sh: 2: cd: can't cd to ../../tests
> >> /bin/sh: 3: cannot open doRUnit.R: No such file
> >> make: *** [test] Error 2
> >>
> >>
> >> As you can see the file doRUnit.R is missing.  It would be great if you
> >> could include this file into the source diustribution to make sure we
> >> can reproduce your exact test procedure in the Debian package.
> >
> > I cloned https://github.com/cran/GenABEL.git and had a look into the
> > code.  I realised that the last tag containing the file doRUnit.R was
> > 1.6-7.  I wonder how you are doing unit testing in the current version.
> >
> > Thanks for any enlightenment
> >
> >      Andreas.
> >
> > --
> > http://fam-tille.de
> 

-- 
http://fam-tille.de